Time Series-Based Predictive Modeling of PM2.5 Levels in Chiang Mai, Thailand

Abstract

The purpose of this study is to use a variety of models to create prediction models for Chiang Mai's PM2.5 levels. To improve the accuracy of our predictions, we take into account outside variables that might influence PM2.5 levels. Among the variables that we include in the data are PM2.5 concentrations, temperature, wind speed, precipitation, cloud cover, relative humidity, and other external factors. Before using the model, the researcher used basic statistical analysis, seasonal analysis, and stationary analysis to assess the data. The team of researchers carried out both data transformation and data cleansing. We tested the ARIMA, SARIMA, and SARIMAX forecasting models. First, we use ARIMA to forecast and assess results. The SARIMA model more accurately captured the seasonal connection in the data when we included a seasonal component. The model was able to forecast PM2.5 levels more precisely at times when seasonal patterns recurred thanks to this improvement. As the last step, we used the SARIMAX model to improve performance by adding exogenous variables. In the end, we assessed the accuracy and performance of each forecast using the MAE and RMSE numbers. The ARIMA model yielded MAE values of 7.34 and RMSE 7.95. The SARIMA model MAE values of 5.76 and RMSE 6.54. The SARIMAX model, when incorporating humidity, had the lowest MAE values of 4.36 and RMSE 5.25, representing improvements MAE of 40.6% and RMSE 34% compared to ARIMA.